what-when-how
In Depth Tutorials and Information
6
5.5
5
4.5
Unchoked
Unchoked (random)
Download
Download (random)
Interest
Interest (random)
Connection
Connection (random)
4
3.5
3
2.5
2
1.5
1
0
5
10
15
20
25
30
35
40
45
50
Time (hours)
Figure6.15
Thecharacteristicpathlengthduringtheexperiments.Alsoshown
arethoseofasimilar-sizedrandomgraph.
Figure 6.16 shows the clustering coefficients of the four networks in the experi-
ment. Although not shown in the figure, the coefficient starts at 1 (since it is a
clique), and then has a sharp decline during the startup stage as the size of the
graph increases. Once all the peers have joined the system there is some further
decrease in the coefficients of all but the Unchoked graph during the transient
stage. hrough the end of the transient stage there are some further small oscilla-
tions in the Interest and Download graphs, until all settled into a steady stage after
approximately 20 hours.
Although at first it seems that there is some clustering present in Figure 6.16,
especially in the graphs of Connection, Interest, and Download peers, further inves-
tigation shows that is not the case. Figure 6.17 shows the clustering coefficients of
the graphs when compared with (divided by) that of a similar sized random graph
(same node and edge restrictions), which is not expected to have any clustering at
all. Here we see that there is some clustering during the startup stage which begins
to decrease once all the nodes have joined the system. he Unchoked graph has no
clustering through the rest of the experiment, while the clustering of the other graphs
reduces more slowly through the transient stage. In the steady stage, all graphs have
almost no clustering. he increased noise in the comparison of the Unchoked graph
with a random graph is due to its relatively tiny clustering coefficient.
Search WWH ::




Custom Search